Eh, usually less than you would expect. We're really good at math and are quite capable of making synthetic experiments where we find people who either require the procedure, or where it's been done incidentally and then inferring the results as though deliberate.

We can also develop a framework for showing benefit from the intervention, perform the intervention ethically, and then compare that to people who didn't get the intervention after the fact. With proper math you can construct the same confidence as a proper study without denying treatment or intentionally inflicting harm.

It's how we have evidence that tooth brushing is good for you. It would be unethical to do a study where we believe we're intentionally inflicting permeant dental damage to people by telling them not to brush for an extended period, but we can find people who don't and look at them.